The effectiveness of externam beam radiation treatment for prostate cancer is decreased due to a variety of uncertainties in the treatment setup, including the physical characteristics of the treatment beam, patient positioning issues, patient organ motion and operator non-reproducibility. The development and administration of a treatment plan using image- guided techniques to account for some of these uncertainties can positively impact its effectiveness. However, the use of these techniques to date has been limited by i.) a lack of accuracy, robustness and reproducibility in the registration of the high resolution 3D computed tomographic (3DCT) or simulator images acquired in a reference )or planning) frame to the highly noisy and blurry portal images, acquired in the treatment environment and ii.) the difficulty in measuring organ motion and relating it to these data. Thus, we first propose to develop a new automated, accurate, and robust system for performing bony anatomy- based 3DCT- to- multiple- (2D) portal image registration by simultaneously incorporating portal image segmentation. The system will rely on a combination of dense field (region-based) and sparse field (gradient/boundary features) information and will use information- theoretic metrics in an optimization framework to solve for the mapping parameters. This approach will be validated using a gold standard developed from serial CT acquisitions taken each week during the treatment. Next, we will study the relationship between setup variation due to bony structure movement and that due to organ motion in preparation for the design of a future complete system that can acquire treatment- environment images of the prostate using an ultrasound probe attached to an articulated arm in an external- skin- marker-based frame, and the 3DCT-to-multiple portal registration algorithm described above. The feasibility of using external markers to relate portal and ultrasound information will be a key part of this study as well. Finally, we will evaluate the utility of the 3DCT-to-multiple portal registration approach by applying it to the problem of quantitatively studying the sensitivity of errors in the delivery of an optimal dose distribution for a particular patient on a particular day to variations in patient- positioning-related setup for treatment plans of different complexity. These studies will help us understand the utility of more complex treatment plans and planning systems in today's health care environment.